What does the end-to-end pipeline look like?

Five stages: ingest, extract, validate, store, and reconcile. Ingest watches an inbox or folder. Extract sends each document through Dots AI against your invoice schema. Validate checks totals and dates. Store writes the JSON to a database. Reconcile matches the invoice against a purchase order. Most teams already have stages one through three partly done; the API handles the tedious middle.

How do you connect it to a workflow tool?

1. Create a webhook that fires when a new email with a PDF attachment arrives.
2. Send the file to the Dots AI extract endpoint with your schema ID.
3. Parse the JSON response for total, vendor, invoice number, and due date.
4. Run a validation rule: does line-item sum equal the stated total?
5. If it passes, create the record in your accounting API; if not, flag for a human.

Tools like n8n, Make, or a short Python script can all drive this. The Dots AI call is a single HTTP request either way.

How do you keep errors out of the ledger?

• Check field confidence: send anything below your threshold to review.
• Enforce math: reject invoices whose line items do not sum to the total.
• De-duplicate: hash the file and invoice number to catch repeats.
• Log raw responses: keep the original JSON for audit trails.

What does it cost at scale?

Pricing is per document, so a company processing a few thousand invoices a month should model the cost against the person-hours it replaces. Include the human-review minutes for flagged documents — usually 5-10% of the volume — in that math, because that is where the real cost sits.

Where does automation still need a human?

Multi-vendor layouts, handwritten notes, and disputes all need eyes. The goal is not zero touch; it is moving people from data entry to exception handling. Start with your cleanest recurring vendor and expand once the pipeline holds.

Quick Summary

• A Dots AI invoice pipeline runs ingest, extract, validate, store, reconcile.
• One HTTP call per document against your saved schema.
• Validate line-item math and de-duplicate by file hash and invoice number.
• Expect 5-10% of documents to need human review.
• Automate the clean vendors first, then expand.

What is Dots AI?

Dots AI is an extraction platform that takes text, PDFs, or documents and returns structured output against a schema you define. You specify fields like invoice_number, total, due_date, and it fills them in. It exposes a REST API and a dashboard. Pricing is usage-based, with a free allowance for testing and per-document pricing at volume.

How do you define an extraction schema?

1. Create a project and name the document type (e.g. “Vendor Invoice”).
2. Add fields with a name and type: string, number, date, currency.
3. Mark which fields are required.
4. Upload a sample document to test.
5. Fix any misses by adding a short description to the field, like “total amount including tax”.

How accurate is it on real documents?

On clean, machine-generated PDFs accuracy is high — fields fill correctly on the first pass. On scanned or faxed documents quality drops, so pair it with OCR upstream. Multi-page invoices where the total is on a different page than the header can confuse the mapping; splitting by section helps. Always keep a confidence score if the API returns one and route low-confidence records to human review.

How do you use Dots AI through the API?

Post your document to the extraction endpoint with your schema ID, then read the JSON response. A typical request sends the file plus the schema reference and returns each field with its value and confidence. Batch endpoints let you queue many documents. For a pipeline, wrap the call in retry logic — network hiccups, not the model, cause most failures.

Where does it stumble?

Handwriting is unreliable. Tables with merged cells often misalign. And the model will occasionally invent a plausible value for a blank field unless you mark the field as nullable, so always allow empty responses for optional data. For high-volume structured extraction from digital documents, Dots AI removes a lot of custom parsing code.

Quick Summary

• Dots AI extracts structured JSON from text, PDFs, and documents.
• You define a schema of named, typed fields, then send documents to the API.
• Accuracy is high on clean digital PDFs, lower on scans and handwriting.
• Mark optional fields nullable so it does not invent values.
• Route low-confidence results to human review.